I just watched a short video of a robot (well, a laptop with a webcam in it – but that’s as much of a robot as anything in The Jetsons is, technically…disappointingly) and it was just recognising objects. We see what the camera sees, accompanied by a short list of possible things whatever it’s looking at could be, with percentage markers next to each word, presumably giving the AI’s certainty on what it thinks it’s looking at. So, it looks at a brush and then reckons it’s 12% sure it’s a dust-pan brush, but then again it calculates that there’s a 7% chance it could be something else, like a really big tooth-brush or even some sort of weird hair brush for a horse. If you add all the percentages up, they only come to about 50%, so I’m guessing there’s another unlisted amount of recognition possibilities this thing has that it doesn’t feel are worth suggesting because they’re too out there. Maybe it’s embarrased? Perhaps there’s a 1.3% chance that it perceives the brush as a hairy little animal, hitherto undiscovered to be alive by humans, that is actually sentient and conscious but just freezes up whenever it’s picked up by humans. I mean, we see dustpans and brushes on the shelves of hardware stores and we just assume that they’re manufactured by a person and placed there to be sold, but is it not possible that actually they aren’t what they seem? Perhaps that’s just where they’ve chosen to live. Maybe they’re conscious, but just can’t move, or let us know. Like stones, or plants might be (stones definitely are, don’t step on them)
Sometimes, this AI seems to be wrong. It thinks that a fridge is actually a whiteboard. But then again, a fridge could be a whiteboard. It’s white, and if you write on it with a whiteboard marker I’m at least 12.6% sure it would work. I’m 16% sure it would wipe clean again afterwards, but then again I’m also 0.9% sure that it could be a coffin or 0.3% sure that it’s a door into a parallel universe. But for the most part, is astoundingly accurate. It can identify a Fern tree way better than I ever could, and it seems to even consider if an indoor plant is artificial or not. Although, all indoor plants are artificial in a way. I find the idea of the consideration of all other possibilities interesting though. This AI hasn’t been around as long as we (you’re human, I’m assuming [26.8%]) and maybe because of that fact, it’s more open minded. It thinks analytically, but doesn’t discount the other possibilities. It’s fallible, or at least seems like it is. The only thing is, it’s learning what the world is based on information fed to it by us. I don’t think it’s making it’s mind up based on it’s own experiences (but it might be, I dunno) but if it isn’t, then it’s learning through our perceptions. And how reliable are they? We only catagorise objects based on our own personal learnt experience, from our lives around us. Our minds have seemingly evolved to understand what things are, but can I really ever trust myself? All I know for certain now is, I will watch my back while walking down the dustpan aisle at the hardware store.
Video credit: “ 5000 categories visual recognition, AI, robotics” by Youtube user: Eugenio Culurciello